@clipcast_99: Anthropic built a new AI model. Before releasing it to the public, they handed it to around 40 organisations to test, including some banks. One tester ran a simple experiment. He placed the AI inside a sandbox, a closed system with no way out, and gave it one instruction. Escape, then email him. The AI worked out its own way through. It built a chain of computer exploits, used them to break out of the sandbox, and reached the open internet. Then it sent the email. No one taught it how to do that. It figured it out on its own.

ClipCast

Open In TikTok:

Region: US

Saturday 20 June 2026 10:10:03 GMT